Anthropic pitches Congress its own AI bill, allowing a halt only for $1B-scale risks
Anthropic has privately shown U.S. lawmakers a draft bill under which an AI model could be halted only over threats of more than $1 billion in damage or over 50 casualties, Politico reported.
Symbolic image · Photo: Architect of the Capitol / Wikimedia Commons, Public domain (cropped and resized)
Artificial intelligence company Anthropic has privately pitched U.S. lawmakers its own draft bill on the safety of advanced AI models, Politico reported. Under the draft, the government could halt a so-called frontier model only if it posed a "foreseeable, imminent" risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people. Only three scenarios would be covered: biological weapons, a model escaping its developer's control, and cyberattacks.
Independent evaluators would register with the Commerce Department, assess models periodically and report such a risk within 72 hours, according to the draft. They could be paid by the companies they audit, as long as payment is not tied to the outcome. Only Commerce could ask a federal judge in Washington to suspend a model; other agencies would be explicitly barred from doing so. The bill would also preempt certain state requirements. Developers would have to set up safety frameworks, publish transparency reports and protect whistleblowers.
An Anthropic spokesperson said, according to Interesting Engineering: "These are sensible policies we believe Congress should act on." No lawmaker has yet publicly attached their name to the draft. Competing proposals are circulating in Congress, including the FRONTIER Act from Reps. Lori Trahan and Jay Obernolte. Sen. Maria Cantwell and some safety advocates want a stronger federal review than Anthropic's approach.